WhiteAiZ's picture
Upload 1420 files
0070fce verified
import { unescapeHtml } from '.'
// Fork from https://github.com/jiw0220/stable-diffusion-image-metadata/blob/main/src/index.ts
type ImageMeta = {
prompt?: string;
negativePrompt?: string;
steps?: string;
sampler?: string;
cfgScale?: string;
seed?: string;
clipSkip?: string;
hashes?: { [k: string]: any };
width?: number;
height?: number;
resources?: Resource[];
} & Record<string, any>;
type Resource = {
type: string;
name: string;
weight?: number;
hash?: string;
};
type PreProcessValueFn = (v: string) => string;
type PreProcessValue = string;
const imageMetadataKeys: Array<[string, string]> = [
['Seed', 'seed'],
['CFG scale', 'cfgScale'],
['Sampler', 'sampler'],
['Steps', 'steps'],
['Clip skip', 'clipSkip'],
['Size', 'size'],
];
const imageMetaKeyMap = new Map<string, string>(imageMetadataKeys);
const automaticExtraNetsRegex = /<(lora|hypernet):([a-zA-Z0-9_.]+):([0-9.]+)>/g;
const automaticNameHash = /([a-zA-Z0-9_.]+)\(([a-zA-Z0-9]+)\)/;
const getImageMetaKey = (key: string, keyMap: Map<string, string>) => keyMap.get(key.trim()) ?? key.trim();
const stripKeys = ['Template: ', 'Negative Template: '] as const;
function preproccessFormatJSONValueFn(v: string) {
try {
return JSON.parse(encodeURIComponent(v));
} catch (e) {
return v;
}
}
function preproccessFormatHandler(configValue: PreProcessValue | PreProcessValueFn, inputValue: string) {
if (typeof configValue === 'function') {
return configValue.call(null, inputValue);
}
return configValue;
}
const tryParseJson = (v: string) => {
try {
return JSON.parse(v);
} catch (e) {
return v;
}
}
const preproccessConfigs = [
{ reg: /(ControlNet \d+): "([^"]+)"/g },
{ reg: /(Lora hashes): "([^"]+)"/g },
{ reg: /(Hashes): ({[^}]+})/g, key: 'hashes', value: preproccessFormatJSONValueFn },
//...There should be many configs that need to be preprocessed in the future
];
export function parse(parameters: string): ImageMeta {
const metadata: ImageMeta = {};
if (!parameters) return metadata;
const metaLines = parameters.split('\n').filter((line) => {
return line.trim() !== '' && !stripKeys.some((key) => line.startsWith(key));
});
const detailsLineIndex = metaLines.findIndex((line) => line.startsWith('Steps: '));
let detailsLine = metaLines[detailsLineIndex] || '';
// Strip it from the meta lines
if (detailsLineIndex > -1) metaLines.splice(detailsLineIndex, 1);
// Remove meta keys I wish I hadn't made... :(
detailsLine = unescapeHtml(detailsLine)
const preprecessedMatchValuesList = [] as any[];
preproccessConfigs.forEach(({ reg, key: configKey, value: configValue }) => {
const matchData: any = {};
const matchValues = [];
let match;
while ((match = reg.exec(detailsLine)) !== null) {
const key = configKey !== void 0 ? preproccessFormatHandler(configKey, match[1]) : match[1];
const value = configValue !== void 0 ? preproccessFormatHandler(configValue, match[2]) : match[2];
matchData[key] = value;
matchValues.push(match[0]);
}
matchValues.forEach((value) => (detailsLine = detailsLine.replace(value, '')));
preprecessedMatchValuesList.push(matchData);
});
const regex = /\s*([\w ]+):\s*("(?:\\"[^,]|\\"|\\|[^"])+"|[^,]*)(?:,|$)/g;
let match;
while ((match = regex.exec(detailsLine)) !== null) {
let k = match[1];
const v = match[2].replace(/\\(.)/g, '$1');
if (!k) continue;
k = getImageMetaKey(k, imageMetaKeyMap);
metadata[k.trim()] = tryParseJson((v ?? '').trim());
}
// 这些信息不是很重要,所以推后
preprecessedMatchValuesList.forEach((matchData) => {
Object.assign(metadata, matchData);
});
// Extract prompts
const [prompt, ...negativePrompt] = metaLines
.join('\n')
.split('Negative prompt:')
.map((x) => x.trim());
metadata.prompt = prompt;
metadata.negativePrompt = negativePrompt.join(' ').trim();
// Extract resources
const extranets = [...prompt.matchAll(automaticExtraNetsRegex)];
const resources: Resource[] = extranets.map(([, type, name, weight]) => ({
type,
name,
weight: parseFloat(weight),
}));
if (metadata.Size || metadata.size) {
const sizes = (metadata.Size || metadata.size || '0x0').split('x');
if (!metadata.width) {
metadata.width = parseFloat(sizes[0]) || 0;
}
if (!metadata.height) {
metadata.height = parseFloat(sizes[1]) || 0;
}
}
if (metadata['Model'] && metadata['Model hash']) {
const model = metadata['Model'] as string;
const modelHash = metadata['Model hash'] as string;
if (typeof metadata.hashes !== 'object') metadata.hashes = {};
if (!metadata.hashes['model']) metadata.hashes['model'] = modelHash;
resources.push({
type: 'model',
name: model,
hash: modelHash,
});
}
if (metadata['Hypernet'] && metadata['Hypernet strength'])
resources.push({
type: 'hypernet',
name: metadata['Hypernet'] as string,
weight: parseFloat(metadata['Hypernet strength'] as string),
});
if (metadata['AddNet Enabled'] === 'True') {
let i = 1;
// eslint-disable-next-line no-constant-condition
while (true) {
const fullname = metadata[`AddNet Model ${i}`] as string;
if (!fullname) break;
const [, name, hash] = fullname.match(automaticNameHash) ?? [];
resources.push({
type: (metadata[`AddNet Module ${i}`] as string).toLowerCase(),
name,
hash,
weight: parseFloat(metadata[`AddNet Weight ${i}`] as string),
});
i++;
}
}
metadata.resources = resources;
return metadata;
}